Agent manager
Claude Code dashboard: the six things you need to see at a glance
Updated November 17, 20267 min read
A Claude Code dashboard is one screen that shows every coding agent you have running, so you can see who is busy, who is stuck and how much of your plan is left without clicking through windows. This page breaks a good dashboard into six parts and, for each one, shows how to get it by hand with Claude Code's own commands and what a ready-made agent manager does instead.
In short
- A useful coding agent dashboard answers six questions: status, waiting questions, usage, context, changes and history.
- Claude Code covers each one with separate commands such as
claude agents,/usage,/contextandgit diff. - The pain starts when you run several agents: the answers live in different windows.
- An agent manager like VibeCodingDeck puts all six on one screen, for Claude Code and other agents.
Six panels every coding agent dashboard needs
With one agent, the terminal tells you everything. With five, you start missing things: an agent that sat idle for twenty minutes waiting for a "yes", or a chat that ran out of room halfway through a job. These six panels are what you want to monitor Claude Code sessions properly:
Status
Is each agent working, idle, waiting or done?
Waiting questions
Which agent is blocked on a permission prompt or a question for you?
Usage
How much of your daily and weekly plan limit is gone, per account?
Context
How full is each chat, so you can clear it before it runs out of room?
Changes
Which files did each agent edit, and is the work safe to merge?
History
Can you reopen yesterday's sessions with their full conversation?
Below, each panel gets its own section: first the do-it-yourself way, then how a dashboard app handles it.
Status: who is working and who has stopped
By hand. Claude Code has its own status screen called agent view. As of November 2026 it is a research preview (an early feature that may change). It lists your Claude Code sessions and shows which are working, need input or are done. Sessions can run in the background with no terminal window open:
claude --bg --name "checkout-bug" "fix the failing checkout tests"
claude agents # open agent view
claude attach checkout-bug # jump into one sessionInside agent view, Enter opens a session, Space shows a quick peek, and Ctrl+X stops one. Some newer terminals can also show each session's state in the tab. That needs Claude Code v2.1.295 or later and a terminal that supports the signal.
In a dashboard. Status is shown as a colour on every pane, all the time, with no command to run. The difference matters most when your agents are a mix of Claude Code and other tools, because agent view only lists Claude sessions.
Waiting questions: the agent that needs a yes
By hand. An agent waiting for permission does nothing until you answer, so this is the panel that saves the most time. Claude Code fires a Notification hook when it is waiting. You can filter it with the permission_prompt matcher for approvals, or agent_needs_input for background sessions while agent view is open. Each hook can play a sound or show a pop-up. Our Claude Code notifications page has ready recipes for Mac, Windows and Linux.
The weak spot: every session plays the same sound, so you hear it and still have to hunt for the window.
In a dashboard. The waiting agent is marked on screen, alerts use different sounds for different levels, and one click takes you to the agent that asked.
Usage: how much of your plan is left
By hand. Type /usage in a session. Subscribers see plan usage bars and a breakdown of what used them; API users see tokens and an estimated cost. The open-source tool ccusage reads your local logs for daily or monthly reports with npx ccusage@latest. Both show one account at a time, and only what happened on this computer. More detail is on our Claude Code usage tracker page.
In a dashboard. A usage meter stays visible, with daily and weekly windows for every account you use. That lets you check you have room before you start six agents on a big job, instead of finding out when they all stop at once.
Context: how full each chat is
By hand. Every chat has a context window, which is the amount of text the AI can hold in mind at once. When it fills up, Claude has to squeeze older parts of the chat and can lose details. /context shows what is filling it, and a custom status line can show context use at the bottom of the screen all the time. /clear starts the chat fresh.
In a dashboard. A small meter on each pane shows how full that chat is, so you can spot the one that needs clearing before it starts forgetting the plan.
Changes: what each agent actually edited
By hand. "Done" only means the agent stopped. You still need to see what it changed. In the project folder, these git commands show it:
git status # which files changed
git diff # the exact lines added and removed
git log --oneline # recent saved changesIf several agents share one project, give each its own git worktree (a second copy of the project folder that git keeps linked) so their edits cannot collide. claude --worktree <name> does this for you. See Claude Code worktrees.
In a dashboard. Each agent works in its own safe copy, and you review and merge the good work back when you are happy with it.
History: picking up where you left off
By hand. claude --continue reopens the last conversation in the current folder, and claude --resume (or /resume inside a session) lets you pick an older one from a list. It works well for one project at a time. With eight sessions across four projects, you end up reopening each one in its own window.
In a dashboard. You close the app, and when you open it again the layout and the real conversations come back in the same panes.
A morning with four agents, with and without one screen
Picture four agents: one fixing a checkout bug, one writing tests, one building a settings page in a second project, and one researching a payment library. Without a dashboard, you cycle through four windows every few minutes. You find the test agent has been waiting on a permission prompt for a quarter of an hour. The research agent filled its chat and started repeating itself. You only learn you are near your weekly limit when two agents stop at once. None of these is hard to fix. The cost is that you only find out late.
With all six panels on one screen, the same morning looks different. The waiting agent shows up as a coloured light and a sound the moment it asks. The full chat shows on its context meter, so you clear it before it drifts. The usage meter tells you to move the research job to a second account before the limit hits. You spend your time reviewing finished work instead of hunting for it.
Should you build your own agent dashboard?
You can glue together tmux, a few hooks, a status line and agent view into your own setup. It can work well. Here is the honest trade:
Building your own
- Free, apart from your time.
- You choose every key and every sound.
- Agent view and hooks are official Claude Code features.
- Fits neatly if you already live in tmux on Mac or Linux.
What it costs you
- Each of the six panels needs its own setup, and they live in different places.
- Agent view only covers Claude Code, so Codex or Gemini need a separate setup.
- Usage across several accounts is hard to see in one view.
- tmux on Windows means running WSL.
- Every new computer needs the same setup again.
Ready-made Claude Code managers
| Tool | Kind | Worth knowing |
|---|---|---|
| Claude Code agent view | Terminal screen | Official, Claude only, research preview |
| Claude desktop app (Code tab) | Desktop app | Sidebar of parallel Claude sessions; see Claude Code desktop |
| Claude Squad | Terminal app | Open source, uses tmux and worktrees; see Claude Squad alternatives |
| Conductor | Mac app | Runs Claude Code, Codex and Cursor agents in isolated workspaces; see Conductor alternatives |
| VibeCodingDeck | Desktop app | Windows and Apple silicon Macs, up to 12 agents per grid, 8 agent types |
All six panels in one VibeCodingDeck window
VibeCodingDeck is an AI agent manager for Windows 10/11 and Macs with Apple silicon. It runs the agents already on your computer (Claude Code, Codex, Gemini CLI, Grok, GitHub Copilot, Qwen, Kimi and Antigravity) and uses your own subscriptions. Mapped to the six panels:
- Status: a live light on every pane (working, idle, waiting, done), up to 12 panes per grid and up to 6 grids one click apart.
- Waiting questions: three alert levels with three different sounds, a jump to the agent that is waiting, and one alert when the whole fleet is done.
- Usage: a live meter with daily and weekly windows for every account, including several logins per AI.
- Context: a context meter on each pane.
- Changes: safe copies per agent, save points you can roll back to, and a merge when you are happy.
- History: sessions resume with the real conversations, and you can search every terminal at once.
It costs $29/month or $249/year as of November 2026, with every feature on every plan. See all features.
Dashboard questions
Does Claude Code have a built-in dashboard?
+
Yes, a basic one. claude agents opens agent view, a terminal screen listing your Claude Code sessions and whether each is working, waiting or done. As of November 2026 it is a research preview.
What is an AI agent manager?
+
An app that shows all your running AI agents on one screen, with their status, usage and project, so you can see which one needs you and switch to it in one click.
Can one dashboard show Claude Code and Codex together?
+
Agent view only lists Claude Code sessions. VibeCodingDeck shows Claude Code, Codex, Gemini CLI, Grok and four other agents side by side in one grid.
Is there a team usage dashboard for Claude Code?
+
Yes, for admins. Claude for Teams and Enterprise has an analytics dashboard, and API organisations get usage reports in the Claude Console. Those track cost and adoption, not live agents.
Sources
Run up to 12 AI coding agents in one window.
Claude Code, Codex, Gemini, Grok and more, side by side. Windows and Mac. From $29 a month.
Get startedKeep reading
Parallel agents
How to run multiple Claude Code agents at once: four ways, easiest last
Usage and limits
Claude Code usage and cost: what you pay, how to check it, how to use less
Alert recipes
Claude Code notifications: copy-paste recipes for every computer
Task boards
Claude Code kanban: a real task board for AI agents, column by column
Mac and Windows
Conductor app on Mac, and the same idea on Windows